DeepSeekAI ModelsBenchmarksAI CodingLLM Pricing

DeepSeek V4 Flash 0731: Current API Alias and Historical Context

DeepSeek's API now serves legacy V4 Flash aliases as V4.1-Flash. See the current context limits and peak/off-peak prices, and why old snapshot benchmarks need separate evidence.

DeepSeek V4 Flash 0731 Architecture and Benchmark Performance Overview
Analyzed Foundation Models (3)Full LLM Leaderboard→
DeepSeek-V4-Flash-0731
DeepSeek
46.1SCORE
DeepSeek-V4-Pro-0813
DeepSeek
54.4SCORE
GLM-5.3
Zhipu AI
56.2SCORE

The identifier “DeepSeek V4 Flash 0731” refers to a historical model snapshot. DeepSeek’s current API documentation maps the legacy deepseek-v4-flash name to DeepSeek-V4.1-Flash; requests using that alias are served by the current model and billed at its current rates. Historical snapshot results and the current API offering should therefore be treated separately.

Current API facts

DeepSeek documents a 1M context length and a maximum output of 384K tokens for its current Flash and Pro API models. Pricing is time-dependent:

Current model IDCache hit input / 1MCache miss input / 1MOutput / 1M
deepseek-flash (V4.1-Flash)$0.003 off-peak; $0.006 peak$0.15 off-peak; $0.30 peak$0.60 off-peak; $1.20 peak
deepseek-v4-pro (V4-Pro-0813)$0.022 off-peak; $0.044 peak$0.66 off-peak; $1.32 peak$1.98 off-peak; $3.96 peak

These are API rates from DeepSeek’s current pricing documentation, which defines separate peak and off-peak prices. Check the source before deploying because rates and model aliases can change.

Architecture claims need a model boundary

DeepSeek’s V4 research paper describes the model family and its long-context design. It reports that V4-Pro uses 27% of the single-token inference FLOPs and 10% of the KV cache of V3.2 in a one-million-token context setting. Those measurements are specific to Pro versus V3.2; they should not be generalized to every V4 model or compared with unrelated systems without matching conditions.

The paper’s family-level architecture discussion covers Compressed Sparse Attention, Heavily Compressed Attention, and Manifold-Constrained Hyper-Connections. The paper does not establish benchmark scores, prices, or maximum output for a July Flash snapshot. Nor should DeepSWE scores be labeled as SWE-bench Verified: they are different evaluations.

Reading benchmark results

The technical paper cited here does not substantiate benchmark results for the “preview” and “0731” snapshots. No comparable score table is included because the public sources linked here do not establish matching evaluation versions, harnesses, and budgets for those snapshots.

For a fair comparison, use results that identify the exact model snapshot, benchmark version, agent scaffold, tool permissions, and attempt budget. Treat provider-reported scores as vendor evaluations unless an independent reproduction is available.

Before integrating, read the current DeepSeek pricing and model documentation and use the model IDs it lists. Our leaderboard may contain separate benchmark records; consult each score’s evaluation source and version.

Frequently Asked Questions

What is DeepSeek V4 Flash 0731?

The 0731 label refers to an earlier model snapshot. DeepSeek's current API documentation says the legacy deepseek-v4-flash alias is served by DeepSeek-V4.1-Flash.

How much does DeepSeek V4 Flash 0731 cost?

DeepSeek currently lists V4.1-Flash at $0.15/$0.30 per million cache-miss input tokens off-peak/peak, $0.003/$0.006 cached input, and $0.60/$1.20 output.

What are the primary benchmark gains over the preview build?

The V4 research paper cited here does not verify the 0731 snapshot benchmark figures. Compare only results that identify the exact snapshot, benchmark version, harness, and budget.

What is the context window of DeepSeek V4 Flash 0731?

DeepSeek's current API documentation lists a 1M context length and maximum output of 384K tokens for its current Flash and Pro models.

Sources & disclosures

Source links for data and claims discussed in this article are listed below. Refer to each source for its current figures and methodology.

Share Article

Recommended Reading

Related by topic and recency.

All articles→
BenchmarksDeepSeek

DeepSeek V4 Pro 0813: Architecture, API Pricing, and Deployment Limits

DeepSeek's V4 Pro report describes a 1.6T-parameter MoE model with 49B active parameters. Current API pricing varies by peak and off-peak periods; see the rate card and deployment caveats.

Aug 13, 2026 · 4 min read
ReviewsDeepSeek

DeepSeek V4 Flash: Current API Name, Pricing, and Model Size

DeepSeek's API now serves V4.1-Flash through the deepseek-flash name. Review the current peak and off-peak price schedule, context size, and the limits of active-parameter comparisons.

Aug 10, 2026 · 3 min read
GuidesCoding Agents

Best LLMs for Coding in 2026: How to Compare Them

Choose a coding LLM by matching the benchmark to your workflow, checking the agent harness, and measuring task quality and cost on your own repository.

Sep 4, 2026 · 3 min read